Add checks for conflicts for aliases

[Mingun]

• Check that alias is not the same as name of other field / variant (it still can be the name of owning field / variant)
• Check that aliases are unique, i. e. two different fields / variants does not use the same alias

It would be better, if error points to corresponding alias, but currently alias spans are not stored anywhere, so I left this to another PR.

Fixes #2308, fixes #2551

[Mingun]

@dtolnay, @oli-obk, can you give some feedback?

[oli-obk]

Annoying that we can't crater it first, but this also catches really subtly broken code, so I guess we can Yolo it.

[dtolnay]

I do not think this can be released as is. This is a pretty blatant breaking change in existing working code.

I would recommend emitting something like a unreachable_patterns warning on duplicated names.

[oli-obk]

The code would never have worked correctly. I don't think anyone would ship this for long without users complaining.

If we were shipping such a breaking change in rustc, this would be the equivalent of adding an error for unsound code that we know has almost no real world usage

[dtolnay]

I do not agree that this PR is analogous to rustc emitting a new error on an unsound code pattern. It is analogous to rustc deciding that unreachable patterns are a hard error now, because their previous behavior cannot possibly be what the person who wrote the unreachable pattern intended. For code in which deserializing to the first variant with the same alias was the desired behavior (hence users not complaining) I don't understand why it's better for a new version of serde to break them as opposed to a warning that says to remove the irrelevant alias.